Assessing the influence of uncertainty on pain perception.

2512 

Abstract Submission 

Mégane Lacombe-Thibault¹, Michel-Pierre Coll¹

¹Laval University, Québec, Quebec

Mégane Lacombe-Thibault  
Laval University
Québec, Quebec

Michel-Pierre Coll  
Laval University
Québec, Quebec

According to a learning account of pain, pain should be modulated to optimize learning and control (Seymour, 2019). Accordingly, previous evidence suggests that the subjective perception of pain is increased when pain leads to prediction errors that can be used to learn new information (Coll et al., 2023). Functional magnetic resonance imaging studies (fMRI) have measured brain activity during various cued pain paradigms and have found mediation effects of the anterior insular cortex on expectation of pain and on the integration of expectation and prediction errors of pain in healthy individuals (Strube et al., 2021; Fazeli and Büchel, 2018). Furthermore, pain perception is likely influenced by the uncertainty of the environment (De Berker et al., 2016) since, in uncertain settings, information gained from pain experiences could be more valuable. However, we still know little about the neural mechanisms underlying the modulation of pain perception by uncertainty. The current study aimed at validating a behavioural task in which pain stimuli could be used to learn cue-outcome contingencies under varying degrees of uncertainty. The ultimate goal is to employ this task in assessing the brain mechanisms responsible for how uncertainty modulates pain perception.

50 healthy adults (mean age of 24.38 years old; 28F, 22M) participated in an experimental task (den Ouden et al., 2010) in which they had to learn to predict the associations between high or low tones and high or low painful electric stimuli. At each trial of this task (Figure 1A), a tone was presented and was followed by a painful stimulus. Participants had to indicate if the pain they received was high or low as quickly as possible. The probabilities related to the associations between the tones and painful stimuli changed throughout the task. The associations remained stable for several trials or changed rapidly in order to create periods of low or high uncertainty (Figure 1B). Different reinforcement learning models and a null model were compared to assess which model would best explain the data across participants. We then used trial-wise estimates of prediction errors and various types of uncertainty to attempt to predict changes in pain perception using linear mixed models.

Participants adequately learned to predict upcoming pain intensity, as confirmed by faster response times when the expected intensity was presented compared to unexpected outcomes (Figure 2A). Furthermore, model comparison confirmed that a hierarchical Bayesian learning model (Mathys et al., 2014) taking into account participants' uncertainty was the best model to explain participants' behaviour (Figure 2B), and this model adequately captured participants' changes in expectations and environmental uncertainty according to the changes in contingencies (Figure 2C and 2D). Computational estimates of environmental uncertainty was a significant estimate of pain ratings (B = 2.15 - 95% CI [0.73, 3.6]), but this relationship interacted with stimulus intensity (B = -4.19 - 95% CI [-7.17, -2.46]) so that high pain stimuli were perceived as less intense under high uncertainty (B = -2.11 - 95% CI [-4.98, -0.18]) while low pain stimuli were perceived as more intense under higher uncertainty (B = 1.84 - 95% CI [0.28, 3.40]).

The results obtained in this study suggest participants were able to accurately learn associations between tones and painful stimuli in the task developed and that pain perception was influenced by estimates of prediction errors and uncertainty. The developed task is therefore suitable to be used in future neuroimaging studies to evaluate how modulation of pain by these learning processes is implemented in the brain.

Learning and Memory Other

Bayesian Modeling ²

Perception: Pain and Visceral ¹

Learning

Modeling

Pain

Perception